Appendix: research commission worksheets

The evidence supplier, analyst and action owner should recognise the same decision in the commission. These completed illustrative dispositions use the useful-signal mechanism as a bounded reference. Empirical likelihoods, actual payoffs, legal authority and implementation readiness remain unresolved. Adaptation therefore requires evidence and accountable approvals, rather than copying the hypothetical purchase conclusion. A filled record makes that work inspectable; a blank form or hypothetical result does not complete it.

The first task is to distinguish a decision question from a research topic. A topic such as whether a proposed activity is promising can support almost any study. A decision question identifies the currently available actions, the represented state, the response window and the consequence of choosing incorrectly. The completed charter asks whether to purchase a binary signal for a specified Commit, Stage or Hold menu. That wording makes it possible to challenge a study that delivers an interesting description without supporting the conditional response policy the institution needs.

The charter should also preserve the current counterfactual. If the institution already intends to Commit irrespective of the signal, buying a study cannot be justified by comparing its favourable outcome with Hold. If the current preferred action is Stage, the signal policy must improve on Stage under the declared objective. The fixture provides that explicit benchmark, but an actual institution must establish what would happen without the proposed evidence. A sponsor who disagrees with the current action should challenge the matrix or admissibility before using an inconsistent counterfactual in the study valuation.

The event register fixes what the represented success state means. A commercial outcome, a technical threshold and a regulatory permission can be different events even when a proposal calls all of them success. The illustrative binary state in the app is deliberately abstract; the commissioning worksheet requires a real decision to define its event, horizon, measurement and information set. If a finding is used to update a different event, the numerical posterior no longer follows merely because the same word appears in both reports. The event owner should confirm that bridge explicitly.

The instrument register asks for conditional likelihood support. It is insufficient for a supplier to state an overall accuracy percentage without the intended event, reference standard and use conditions. The app needs a positive-signal likelihood given success and a negative-signal likelihood given failure. For a real commission, those inputs require evidence, uncertainty and transfer limits. The worksheet therefore separates the supplied illustrative likelihoods from the unresolved empirical validation. It can support a scoped mechanism demonstration while refusing to certify that an actual instrument has the fixture's performance.

The payoff record should show why each action cell belongs in one matrix. An action can have an upside in success and a burden in failure, but those amounts need a shared baseline, horizon and institutional incidence. A reversible Stage action may retain benefits and avoid losses differently from a full commitment. The fixture represents these consequences in aggregate; it does not explain a technology, schedule or contract behind the Stage label. The worksheet asks the action owner to supply that explanation and to confirm that the choice can actually be taken in both states.

Study cost is a separate commissioning perimeter. A quoted vendor fee may exclude internal review, operational disruption, interpretation and waiting burdens. Conversely, an operating consequence already inside an action payoff should not be counted again merely because the institution buys evidence. The completed cost record requires a component owner, treatment and timing statement. It declines to populate real amounts. The aggregate fixture cost is sufficient for the illustrative subtraction, while an actual purchase needs its own complete acquisition account and a check against duplication.

The response protocol is the practical link between a posterior and a decision. It should identify the signal format, the action proposed for each branch, the authority needed to implement it and the treatment of an unusable delivery. A result that arrives after an irreversible commitment may have no value for that commitment even if it remains technically informative. A result that requires authority the institution lacks can encounter a similar implementation gap. The worksheet records these as feasibility obligations. It does not hide them in a probability adjustment with no defined empirical meaning.

The acceptance specification should state what the supplier must deliver for the decision use. This might include a defined event, conditional performance evidence, uncertainty description, full data provenance, a method that can be inspected and a timely interpretable output. The actual scope depends on the commission. The completed example requires a binary classification object appropriate to the represented event and a transparent bridge into the model inputs. It does not demand a universal report format. Acceptance should be based on whether the contracted evidence object supports the declared use, not whether the supplier repeats the committee's preferred conclusion.

A challenge worksheet makes the recommendation conditional in a useful way. It identifies a premise whose admissible change could alter the commissioning decision, the evidence needed to resolve it and the interim action. For this example, the response menu, likelihood transfer, cost perimeter and usable window are decisive candidates. The paper's designed counterexamples show different ways these premises matter. An actual challenge should avoid claiming that a chosen stress value is probable unless there is support for that probability. The purpose is to expose dependence on an assumption, then commission or collect the specific evidence that can resolve it.

The minute should record each judgement under its owner: the analyst reproduces the frozen canonical output; the method reviewer checks the stated binary equations; the action owner confirms feasibility and mandate; the evidence owner defends empirical inputs. None supplies the others. Retain unresolved conditions beside the illustrative result so a computational signature cannot later be read as certification of an actual research purchase.

The interim disposition is sometimes the most useful part of the record. If the instrument's performance is unsupported, the institution can ask for a validation proposal, choose a narrower pilot or decline the purchase. If the response is infeasible, it can repair the action menu or change the research purpose. If cost is incomplete, it can request a full acquisition account before comparing a price with information value. These responses turn uncertainty into a concrete work programme. They are more useful than treating every incomplete input as an invitation to select a convenient point estimate and proceed.

The worksheets also help structure a supplier discussion. The client can explain the decision event, intended responses and evidence acceptance conditions without asking the supplier to promise a favourable policy outcome. The supplier can challenge an impossible likelihood request, an unsuitable reference standard or a response window that cannot be met. The analyst can then revise the model perimeter before a commercial commitment. This is a substantive negotiation about what evidence is usable. The binary fixture can clarify incentives and thresholds, but it cannot establish the actual contract terms or resolve a dispute about applicable duties.

After delivery, a closeout record should compare the accepted evidence object with the commissioning purpose. It should preserve the finding, input changes, selected action, authority and material departure from the protocol. A disappointing signal is not automatically supplier failure; an unusable or unsupported delivery is a different question. Likewise, an unfortunate outcome does not by itself prove that a coherent expected-value decision was mistaken. The institution should separate method error, input error, implementation failure and uncertainty that remained within the accepted scope. This classification allows learning without rewriting the original record around hindsight.

Repeated commissions could support a separately designed study of delivery quality, response implementation and calibration. A successful-case archive alone cannot establish representative frequencies. Declare the collection universe, inclusion rules and comparison method before an empirical claim. This paper's illustrations have no such universe; future evaluation may use the original decision records as traceable inputs, but cannot treat them as proof of organisational improvement.

Follow a hypothetical capacity gap through the filled record. An acceptable instrument specification leaves purchase held when the favourable-result response lacks a usable resource reservation. The response owner records missing capacity; the interim minute retains feasible Stage or Hold; the handoff requires capacity work rather than extra likelihood precision. If capacity becomes available, rerun the current menu and acquisition account before progression. The completed charter exposes the gap without supplying authority or field evidence.